R: Pattern Matching and Replacement (original) (raw)

Description

grep, grepv, grepl, regexpr, gregexpr, regexec and gregexec search for matches to argumentpattern within each element of a character vector: they differ in the format of and amount of detail in the results.

sub and gsub perform replacement of the first and all matches respectively within each element of a character vector.

Usage

grep(pattern, x, ignore.case = FALSE, perl = FALSE, value = FALSE,
     fixed = FALSE, useBytes = FALSE, invert = FALSE)

grepv(pattern, x, ignore.case = FALSE, perl = FALSE, value = TRUE,
      fixed = FALSE, useBytes = FALSE, invert = FALSE)

grepl(pattern, x, ignore.case = FALSE, perl = FALSE,
      fixed = FALSE, useBytes = FALSE)

sub(pattern, replacement, x, ignore.case = FALSE, perl = FALSE,
    fixed = FALSE, useBytes = FALSE)

gsub(pattern, replacement, x, ignore.case = FALSE, perl = FALSE,
     fixed = FALSE, useBytes = FALSE)

regexpr(pattern, text, ignore.case = FALSE, perl = FALSE,
        fixed = FALSE, useBytes = FALSE)

gregexpr(pattern, text, ignore.case = FALSE, perl = FALSE,
         fixed = FALSE, useBytes = FALSE)

regexec(pattern, text, ignore.case = FALSE, perl = FALSE,
        fixed = FALSE, useBytes = FALSE)

gregexec(pattern, text, ignore.case = FALSE, perl = FALSE,
        fixed = FALSE, useBytes = FALSE)

Arguments

Details

Arguments which should be character strings or character vectors are coerced to character if possible.

Each of these functions operates in one of three modes:

1. fixed = TRUE: use exact matching.
2. perl = TRUE: use Perl-style regular expressions.
3. fixed = FALSE, perl = FALSE: use POSIX 1003.2 extended regular expressions (the default).

See the help pages on regular expression for details of the different types of regular expressions.

The two *sub functions differ only in that sub replaces only the first occurrence of a pattern whereas gsubreplaces all occurrences. If replacement contains backreferences which are not defined in pattern the result is undefined (but most often the backreference is taken to be "").

For regexpr, gregexpr, regexec and gregexecit is an error for pattern to be NA, otherwise NAis permitted and gives an NA match.

grep and grepv only differ in the default of thevalue argument.

Both grep and grepl take missing values in x as not matching a non-missing pattern.

The main effect of useBytes = TRUE is to avoid errors/warnings about invalid inputs and spurious matches in multibyte locales, but for regexpr it changes the interpretation of the output. It inhibits the conversion of inputs with marked encodings, and is forced if any input is found which is marked as "bytes" (see[Encoding](../../base/help/Encoding.html)).

Caseless matching does not make much sense for bytes in a multibyte locale, and you should expect it only to work for ASCII characters ifuseBytes = TRUE.

regexpr and gregexpr with perl = TRUE allow Python-style named captures, but not for long vector inputs.

Invalid inputs in the current locale are warned about up to 5 times.

Caseless matching with perl = TRUE for non-ASCII characters depends on the PCRE library being compiled with ‘Unicode property support’, which PCRE2 is by default.

Value

grep(value = FALSE) returns a vector of the indices of the elements of x that yielded a match (or not, forinvert = TRUE). This will be an integer vector unless the input is a long vector, when it will be a double vector.

grep(value = TRUE) returns a character vector containing the selected elements of x (after coercion, preserving names but no other attributes).

grepl returns a logical vector (match or not for each element ofx).

sub and gsub return a character vector of the same length and with the same attributes as x (after possible coercion to character). Elements of character vectors x which are not substituted will be returned unchanged (including any declared encoding ifuseBytes = FALSE). If useBytes = FALSE a non-ASCII substituted result will often be in UTF-8 with a marked encoding (e.g., if there is a UTF-8 input, and in a multibyte locale unless fixed = TRUE). Such strings can be re-encoded by [enc2native](../../base/help/enc2native.html). If any of the inputs is marked as "bytes", elements of character vectors x which are substituted will be returned marked as"bytes", but the encoding flag on elements not substituted is unspecified (it may be the original or "bytes"). If none of the inputs is marked as "bytes", but useBytes = TRUE is given explicitly, the encoding flag is unspecified even on the substituted elements (it may be "bytes" or "unknown", possibly invalid in the current encoding). Mixed use of "bytes" and other marked encodings is discouraged, but if still desired one may use [iconv](../../base/help/iconv.html) to re-encode the result e.g. to UTF-8 with suitably substituted invalid bytes.

regexpr returns an integer vector of the same length astext giving the starting position of the first match or-1 if there is none, with attribute "match.length", an integer vector giving the length of the matched text (or -1 for no match). The match positions and lengths are in characters unlessuseBytes = TRUE is used, when they are in bytes (as they are for ASCII-only matching: in either case an attributeuseBytes with value TRUE is set on the result). If named capture is used there are further attributes"capture.start", "capture.length" and"capture.names".

gregexpr returns a list of the same length as text each element of which is of the same form as the return value forregexpr, except that the starting positions of every (disjoint) match are given.

regexec returns a list of the same length as text each element of which is either -1 if there is no match, or a sequence of integers with the starting positions of the match and all substrings corresponding to parenthesized subexpressions ofpattern, with attribute "match.length" a vector giving the lengths of the matches (or -1 for no match). The interpretation of positions and length and the attributes followsregexpr.

gregexec returns the same as regexec, except that to accommodate multiple matches per element of text, the integer sequences for each match are made into columns of a matrix, with one matrix per element of text with matches.

Where matching failed because of resource limits (especially forperl = TRUE) this is regarded as a non-match, usually with a warning.

Warning

The POSIX 1003.2 mode of gsub and gregexpr does not work correctly with repeated word-boundaries (e.g.,pattern = "\b"). Use perl = TRUE for such matches (but that may not work as expected with non-ASCII inputs, as the meaning of ‘word’ is system-dependent).

Performance considerations

If you are doing a lot of regular expression matching, including on very long strings, you will want to consider the options used. Generally perl = TRUE will be faster than the default regular expression engine, and fixed = TRUE faster still (especially when each pattern is matched only a few times).

If you are working with texts with non-ASCII characters, which can be easily turned into ASCII (e.g. by substituting fancy quotes), doing so is likely to improve performance.

If you are working in a single-byte locale (though not common since R 4.2) and have marked UTF-8 strings that are representable in that locale, convert them first as just one UTF-8 string will force all the matching to be done in Unicode, which attracts a penalty of around 3\times{}for the default POSIX 1003.2 mode.

While useBytes = TRUE will improve performance further, because the strings will not be checked before matching and the actual matching will be faster, it can produce unexpected results so is best avoided. Withfixed = TRUE and useBytes = FALSE, optimizations are in place that take advantage of byte-based matching working for such patterns in UTF-8. With useBytes = TRUE, character ranges, wildcards, and other regular expression patterns may produce unexpected results.

PCRE-based matching by default used to put additional effort into ‘studying’ the compiled pattern when x/text has length 10 or more. That study may use the PCRE JIT compiler on platforms where it is available (see [pcre_config](../../base/help/pcre%5Fconfig.html)). As from PCRE2 (PCRE version >= 10.00 as reported by[extSoftVersion](../../base/help/extSoftVersion.html)), there is no study phase, but the patterns are optimized automatically when possible, and PCRE JIT is used when enabled. The details are controlled by[options](../../base/help/options.html) PCRE_study and PCRE_use_JIT. (Some timing comparisons can be seen by running file ‘tests/PCRE.R’ in the R sources (and perhaps installed).) People working with PCRE and very long strings can adjust the maximum size of the JIT stack by setting environment variableR_PCRE_JIT_STACK_MAXSIZE before JIT is used to a value between1 and 1000 in MB: the default is 64. When JIT is not used with PCRE version < 10.30 (that is with PCRE1 and old versions of PCRE2), it might also be wise to set the optionPCRE_limit_recursion.

Warning

An all too common mis-usage is to pass unnamed arguments which are then matched to one or more of ignore.case, perl, value,fixed, useBytes and invert. So it is good practice to name all the arguments (especially asx/text is not the first argument).

Note

Aspects will be platform-dependent as well as locale-dependent: for example the implementation of character classes (except[:digit:] and [:xdigit:]). One can expect results to be consistent for ASCII inputs and when working in UTF-8 mode (when most platforms will use Unicode character tables, although those are updated frequently and subject to some degree of interpretation – is a circled capital letter alphabetic or a symbol?). However, results in 8-bit encodings can differ considerably between platforms, modes and from the UTF-8 versions.

Source

The C code for POSIX-style regular expression matching has changed over the years. As from R 2.10.0 (Oct 2009) the TRE library of Ville Laurikari (https://github.com/laurikari/tre) is used. The POSIX standard does give some room for interpretation, especially in the handling of invalid regular expressions and the collation of character ranges, so the results will have changed slightly over the years.

For Perl-style matching PCRE2 or PCRE (https://www.pcre.org) is used: again the results may depend (slightly) on the version of PCRE in use.

References

Becker, R. A., Chambers, J. M. and Wilks, A. R. (1988)The New S Language. Wadsworth & Brooks/Cole (grep)

See Also

regular expression (aka [regexp](../../base/help/regexp.html)) for the details of the pattern specification.

[regmatches](../../base/help/regmatches.html) for extracting matched substrings based on the results of regexpr, gregexpr and regexec.

[glob2rx](../../utils/html/glob2rx.html) to turn wildcard matches into regular expressions.

[agrep](../../base/help/agrep.html) for approximate matching.

[charmatch](../../base/help/charmatch.html), [pmatch](../../base/help/pmatch.html) for partial matching,[match](../../base/help/match.html) for matching to whole strings,[startsWith](../../base/help/startsWith.html) for matching of initial parts of strings.

[tolower](../../base/help/tolower.html), [toupper](../../base/help/toupper.html) and [chartr](../../base/help/chartr.html)for character translations.

[apropos](../../utils/html/apropos.html) uses regexps and has more examples.

[grepRaw](../../base/help/grepRaw.html) for matching raw vectors.

Options PCRE_limit_recursion, PCRE_study andPCRE_use_JIT.

[extSoftVersion](../../base/help/extSoftVersion.html) for the versions of regex and PCRE libraries in use, [pcre_config](../../base/help/pcre%5Fconfig.html) for more details for PCRE.

Examples

grep("[a-z]", letters)

txt <- c("arm","foot","lefroo", "bafoobar")
if(length(i <- grep("foo", txt)))
   cat("'foo' appears at least once in\n\t", txt, "\n")
i # 2 and 4
txt[i]

## Double all 'a' or 'b's;  "\" must be escaped, i.e., 'doubled'
gsub("([ab])", "\\1_\\1_", "abc and ABC")

txt <- c("The", "licenses", "for", "most", "software", "are",
  "designed", "to", "take", "away", "your", "freedom",
  "to", "share", "and", "change", "it.",
  "", "By", "contrast,", "the", "GNU", "General", "Public", "License",
  "is", "intended", "to", "guarantee", "your", "freedom", "to",
  "share", "and", "change", "free", "software", "--",
  "to", "make", "sure", "the", "software", "is",
  "free", "for", "all", "its", "users")
( i <- grep("[gu]", txt) ) # indices
stopifnot( txt[i] == grep("[gu]", txt, value = TRUE) )

## Note that for some implementations character ranges are
## locale-dependent (but not currently).  Then [b-e] in locales such as
## en_US may include B as the collation order is aAbBcCdDe ...
(ot <- sub("[b-e]",".", txt))
txt[ot != gsub("[b-e]",".", txt)]#- gsub does "global" substitution
## In caseless matching, ranges include both cases:
a <- grep("[b-e]", txt, value = TRUE)
b <- grep("[b-e]", txt, ignore.case = TRUE, value = TRUE)
setdiff(b, a)

txt[gsub("g","#", txt) !=
    gsub("g","#", txt, ignore.case = TRUE)] # the "G" words

regexpr("en", txt)

gregexpr("e", txt)

## Using grepl() for filtering
## Find functions with argument names matching "warn":
findArgs <- function(env, pattern) {
  nms <- ls(envir = as.environment(env))
  nms <- nms[is.na(match(nms, c("F","T")))] # <-- work around "checking hack"
  aa <- sapply(nms, function(.) { o <- get(.)
               if(is.function(o)) names(formals(o)) })
  iw <- sapply(aa, function(a) any(grepl(pattern, a, ignore.case=TRUE)))
  aa[iw]
}
findArgs("package:base", "warn")

## trim trailing white space
str <- "Now is the time      "
sub(" +$", "", str)  ## spaces only
## what is considered 'white space' depends on the locale.
sub("[[:space:]]+$", "", str) ## white space, POSIX-style
## what PCRE considered white space changed in version 8.34: see ?regex
sub("\\s+$", "", str, perl = TRUE) ## PCRE-style white space

## capitalizing
txt <- "a test of capitalizing"
gsub("(\\w)(\\w*)", "\\U\\1\\L\\2", txt, perl=TRUE)
gsub("\\b(\\w)",    "\\U\\1",       txt, perl=TRUE)

txt2 <- "useRs may fly into JFK or laGuardia"
gsub("(\\w)(\\w*)(\\w)", "\\U\\1\\E\\2\\U\\3", txt2, perl=TRUE)
 sub("(\\w)(\\w*)(\\w)", "\\U\\1\\E\\2\\U\\3", txt2, perl=TRUE)

## named capture
notables <- c("  Ben Franklin and Jefferson Davis",
              "\tMillard Fillmore")
# name groups 'first' and 'last'
name.rex <- "(?<first>[[:upper:]][[:lower:]]+) (?<last>[[:upper:]][[:lower:]]+)"
(parsed <- regexpr(name.rex, notables, perl = TRUE))
gregexpr(name.rex, notables, perl = TRUE)[[2]]
parse.one <- function(res, result) {
  m <- do.call(rbind, lapply(seq_along(res), function(i) {
    if(result[i] == -1) return("")
    st <- attr(result, "capture.start")[i, ]
    substring(res[i], st, st + attr(result, "capture.length")[i, ] - 1)
  }))
  colnames(m) <- attr(result, "capture.names")
  m
}
parse.one(notables, parsed)

## Decompose a URL into its components.
## Example by LT (http://www.cs.uiowa.edu/~luke/R/regexp.html).
x <- "http://stat.umn.edu:80/xyz"
m <- regexec("^(([^:]+)://)?([^:/]+)(:([0-9]+))?(/.*)", x)
m
regmatches(x, m)
## Element 3 is the protocol, 4 is the host, 6 is the port, and 7
## is the path.  We can use this to make a function for extracting the
## parts of a URL:
URL_parts <- function(x) {
    m <- regexec("^(([^:]+)://)?([^:/]+)(:([0-9]+))?(/.*)", x)
    parts <- do.call(rbind,
                     lapply(regmatches(x, m), `[`, c(3L, 4L, 6L, 7L)))
    colnames(parts) <- c("protocol","host","port","path")
    parts
}
URL_parts(x)

## gregexec() may match multiple times within a single string.
pattern <- "([[:alpha:]]+)([[:digit:]]+)"
s <- "Test: A1 BC23 DEF456"
m <- gregexec(pattern, s)
m
regmatches(s, m)

## Before gregexec() was implemented, one could emulate it by running
## regexec() on the regmatches obtained via gregexpr().  E.g.:
lapply(regmatches(s, gregexpr(pattern, s)),
       function(e) regmatches(e, regexec(pattern, e)))

[Package _base_ version 4.6.0 Index]